FastEMRIWaveforms: New tools for millihertz gravitational-wave data analysis

Today we released a follow up paper to our Physical Review Letter that presented the first fully relativistic EMRI waveform model fast enough for performing parameter estimation. I outlined the model in a previous blog post. This paper provides a lot more detail and shows the utility of the new model by performing the first Markov Chain Monte Carlo-based EMRI parameter estimation with a relativistic model. We find two key results from these studies: (i) we can significantly reduce the number of harmonic modes in the model without introduction biases in the signal extraction. This helps to speed up the model even further so that, on a GPU, you can compute an EMRI waveform in under 50 milliseconds. And (ii) using semi-relativistic, ‘kludge’ waveform amplitudes introduces significant biases in the posterior distribution for some parts of the parameter space. The below corner plot shows some of the biases that can be introduced when using a Schwarzschild version of the Augmented Analytic Kludge to generate the waveform amplitudes. We gave both inspirals have exactly the same phase from an adiabatic inspiral so the differences are just due to the waveform generation step.


Screenshot 2021-04-13 at 22.05.49

Our present model is only for non-rotating black holes but it is available from a common Python interface, with all the code available from the Black Hole Perturbation Toolkit. In the LISA Data Challenges it is useful to models that cover the full parameter space of generic Kerr orbits (even if they are not phase accurate with the true waveform). For that reason we also provide an updated Augmented Analytic Kludge model via the same common interface. Both waveform models can be GPU accelerated so they generate waveforms in under a second and the common interface means we can seamless replace the kludge model with the fully relativistic model in the future.

There are a lot of nice results in the paper so I recommend you take a look (if EMRI waveforms are your thing). We are (virtual) running the BritGrav meeting at UCD this week and I spent 3 hours chair a session today so I will leave it there for now and get some sleep!